Alibaba Cloud397B (17B active)IQ2_M (sub-Q4) 下约 240 GB 显存

Qwen 3.5 397B-A17B — 显存、速度与本地部署

作者: Jakub Rusinowski · 最后更新:

Alibaba's Feb 16, 2026 flagship MoE model — 397B total, ~17B active. Even at aggressive sub-Q4 quantization it needs ~125 GB (e.g. a 128GB Mac Studio at ~24 tok/s). Not realistically self-hostable on consumer hardware; Ollama only lists a cloud-hosted `:cloud` tag, not a local download. Best treated as a reference/'what frontier open weights look like' entry rather than a local recommendation.

Qwen 3.5 397B-A17B 在 IQ2_M (sub-Q4) 下约需 240 GB 显存——量化权重加框架开销,不含 KV 缓存。在 Apple Silicon 上,这部分来自统一内存。

逻辑96
创意94
编程95

按量化级别的显存与速度

计算基准:NVIDIA RTX 4090 (24 GB)。仅含权重与开销:该模型架构未公开,因此未计入 KV 缓存。

量化显存速度(估算)适配
Q2_K
2.63 bpw
131.3 GB—放不下
Q3_K_M
3.41 bpw
170 GB—放不下
Q4_K_M
4.83 bpw
240.5 GB—放不下
Q5_K_M
5.67 bpw
282.2 GB—放不下
Q6_K
6.56 bpw
326.3 GB—放不下
Q8_0
8.50 bpw
422.6 GB—放不下
F16
16.00 bpw
794.8 GB—放不下

黑色标记 = NVIDIA RTX 4090 (24 GB) 上的可用显存。 估算来自内存带宽屋顶线模型,详见 方法说明页. Qwen 3.5 397B-A17B 显存计算器 →

运行 Qwen 3.5 397B-A17B

目录中能运行 Qwen 3.5 397B-A17B 的最便宜 GPU 是 Apple M3 Ultra (512 GB).

购买此硬件 Apple Mac Studio M3 Ultra — 512 GB VRAM · 60 W board power立即云端部署 RunPod

或在 Vast.ai 比较

作为亚马逊联盟成员,我们从符合条件的购买中获得收入。云 GPU 链接为推荐链接——我们可能获得佣金,您无需额外付费。

联盟营销声明: 本页部分链接为联盟推广链接——如果你通过它们购买,LLM Configurator 可能会获得佣金,而你无需支付任何额外费用。作为亚马逊联盟成员(Amazon Associate),LLM Configurator 会从符合条件的购买中获得收益。
Apple Mac Studio M3 Ultra
512 GB VRAM · 60 W board power
2026年价格波动较大——请以当前商品页价格为准。

如何运行 Qwen 3.5 397B-A17B

The Ollama tag `qwen3.5:397b-cloud` is cloud-hosted, so there is no local run command for this model.

Hugging Face 上的权重: Qwen/Qwen3.5-397B-A17B-Instruct ↗

规格

Curated — A hand-written entry from before this catalogue recorded its sources. The figures are long-standing but their provenance is not on file.

参数量
397 Billion (~17B active)
上下文窗口
262,144
架构
Hybrid Gated DeltaNet + MoE
提供商
Alibaba Cloud
许可证
Apache 2.0
规格量化
IQ2_M (sub-Q4)
系统内存
256 GB
记录更新于
2026-02-16
许可证Apache-2.0允许商业使用

Commercial use permitted. No usage restrictions beyond attribution.

质量与使用场景

评分由模型作者或独立评测方发布——衡量质量而非吞吐量,并非我们实测。

最适合frontier tasksenterpriseresearchcloud api

我的 GPU 能运行 Qwen 3.5 397B-A17B 吗?

Qwen 3.5 的其他尺寸

Qwen 3.5 397B-A17B — 常见问题

How much VRAM does Qwen 3.5 397B-A17B need?

About 240 GB at IQ2_M (sub-Q4) — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.

Does Qwen 3.5 397B-A17B run on an RTX 4090 (24 GB)?

No. Qwen 3.5 397B-A17B needs about 240 GB at IQ2_M (sub-Q4), more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.

How do I run Qwen 3.5 397B-A17B locally?

The Ollama tag `qwen3.5:397b-cloud` is cloud-hosted, so there is no local run command for this model. Running the published tag would send your prompts to a hosted GPU rather than your own machine.

What other sizes does Qwen 3.5 come in?

Qwen 3.5 0.8B (1 GB), Qwen 3.5 2B (2 GB), Qwen 3.5 4B (3 GB), Qwen 3.5 9B (6 GB), Qwen 3.5 27B (17 GB), Qwen 3.5 35B-A3B (22 GB), Qwen 3.5 122B-A10B (74 GB), Qwen 3.5 397B-A17B (240 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.